A new Zaha Hadid Architects-designed showroom for Chinese womens fashion brand Marisfrolg, built with recycled materials, has opened in Shenzhen.Image: MarisfrolgLocated at the heart of the Marisfrolg Fashion Groups headquarters, the roughly 31,215-square-foot showroom interiors was inspired by the design ethos, textiles, and tailoring embodied within the groups nine labels.Image: MarisfrolgThe project includes the main flagship store, a pop-up store for temporary exhibitions and events, a homeware gallery for dcor and furniture, a caf, and an outdoor courtyard.Image: MarisfrolgEchoing the headquarters butterfly-wing structure, the showroom exhibits a unifying architectural language of dynamism, with distinct elements unique to each label. As a result, there are different spatial experiences throughout the site.Image: Zhu YumengThe showroom comprises a variety of recycled aggregates. Its floors feature recycled crushed glass, quartz, sand, and stone. The spaces partitio...

Bandai Namco has launched its Tales remaster project with resounding success. This isnt just some lazy port from Bandai Namco. No, in fact, I was stunned when I booted up the game for the first time with a long list of customizable gameplay options staring down at me. Tales of Graces f has been graced by an impressive collection of new convenience features, to say the least. Most surprising to me is the Grade Shop being accessible right from the start of a new game.The Grade Shop gives players a large selection of mutators that alter the game in various ways, and until now was reserved for new game plus. Want to get battles done with faster? Just toggle the double damage Grade on and prepare for some whirlwind fights involving fast-reflexes. A decent chunk of Grades fall into the handicap category: things like half-price shops, double experience gain, double drop rates, you get the gist. Hardcore players may shake scowl at the fact that every Grade is freely available right from the start, but theres nothing stopping anyone from ignoring it should they yearn for a purist run. Not to mention various Grade options can make the game significantly tougher, like trading all experience for money, for example."The Grade Shop gives the player a large selection of mutators that alter the game in various ways, and its available from the start for the first time."Of course, theres more new here than just a frontloaded and free Grade Shop. Enemy encounters can be toggled on or off, cutscenes and skits are finally skippable, theres destination icons and event indicators, regular battles have a retry option, Japanese voices arrive for the west, and so much more. And it wouldnt be a remaster without some amount of graphical touch-up. While small, the colors and textures have been enhanced for HDR and 4K monitors.Pretty much every convenience feature listed here can be toggled on or off, making the experience personalized to the tiniest detail. Want a vanilla OG experience? Ignore the Grade Shop and toggle all the new features off to have yourself a challenging, yet satisfying playthrough. The secret difficulty setting is still locked behind a completed save file, so that reward has been preserved from the original at least.But why invest yourself in a 15-year-old Tales game in the first place? Well, even A JRPG fan like myself questioned whether Graces held up over the years. I never got around to Graces and didnt know what to expect. Tales of Symphonia was a longtime favorite of mine and while I enjoyed Tales of Arise, it didnt quite have that witty writing and enjoyable plot structure that I loved in Symphonia. Im also fond of Abyss, and to a lesser extent Vesperia, but wasnt floored by either of them. Thats just some context of where Im coming from; someone who likes the series but doesnt quite love it until now.Tales of Graces f has unraveled my JRPG world and Im simply at a loss of how to express my current love affair with it. My expectations were sea-level, and I just got blasted sky-high with what Ive been experiencing. Anyway, onward to the rest of the review."Colors and textures have been enhanced for HDR and 4K monitors. The left side of the image is from the PS3 version, while the right is Remastered."After hearing such praise, youre probably expecting the games opening to rival The Empire Strikes Back or something. Sorry to disappoint, but the first hour of Tales of Graces is a slow burn, not the harmful kind though. The prologue chronicles the cast of characters in their childhood. Children will be children, and this game does a fine job of portraying how kids would behave in this games society. While I thoroughly respect the games childhood chapter, theres no denying the first hour lingers a tad too long on some scenes. Not to mention the child voices grate compared to their adult counterparts. That said, the pacing picks up considerably and glides at a consistent pace after the initial establishment. I implore players to give the narrative a chance. The first hour may seem trope-filled and somewhat dull, but the set-up and follow-through is nothing short of extraordinary. I just cant get over how well the cast is characterized and developed through the roller-coaster of a plot, but more on that later.One way the game characterizes the cast is through fully-voiced Skit events. Skits are a series staple featuring animated character banter ranging from heavy confessions to discussions on the local cuisine of the area. They do a great job characterizing the party members and showing different relationship dynamics.Thankfully, theres a consistent number of Skit events scattered through every town and field area; you never go 30 minutes without seeing a new one. Graces features some strong voice talent too, with Laura Bailey and Cassandra Lee Morris playing a couple of your party members. My only gripe with Skits are the random auto-advance without warning, causing me to skip dialogue occasionally. To the games credit, theres a way to view all 400 and something Skits, but it involves an obscure cat-feeding quest."Skits do a great job characterizing the party members and showing different relationship dynamics."Thats another thing I adore about this game. Theres just so much to do at any given point, and its all so rewarding and fun. Like other Tales games, you can control any party member while exploring on the field and in battle. This game goes further and adds a bunch of unique commentary when examining environmental points of interest with different characters. When Im not exploring each characters reactions to Asbels messy desk or the regional anomaly, Im trying to solve each towns unique treasure puzzle. These locked chests usually require some knowledge about the areas lore to open. Some are particularly clever, like having you look up an NPCs favorite monster and inputting its bestiary number in the chest lock.Theres even a trading card game that you can challenge characters with. Youll pick up cards of popular characters from the Tales series throughout your travels and play a game of guess the quote with NPCs throughout your journey. Its a neat little side game for hardcore series fans, but totally skippable for the rest of us. Speaking of NPCs, they typically give useful tips and tutorials early on, though they mainly supply regional lore and humorous anecdotes. Within each region and town you visit, theres a plethora of Inn requests, local Discoveries, Skits, and time-sensitive events. The game is jam-packed with engaging content every single step of the way."Within each region and town you visit, theres a plethora of Inn requests, local Discoveries, Skits, and time-sensitive events."The real meat and potatoes of the game, though, lies in its amazingly deep combat. This is by far my favorite battle system in the series. Combat seems almost too simple at the start, but it has a depth that Im still exploring 50 hours later. Most of that depth comes from the breathtaking Title system. Titles bestow passive buffs, new skills (or artes as this game terms it), and other unique affects to each character. You unlock Titles with every single action you take in the game. Think of The Elder Scrolls and its all-encompassing use-it-to-level-it-system and you have the right idea. Like to use Demon Fang a lot? Youll no doubt get rewarded with a Title related to that arte.In fact, each party member has, and Im not exaggerating here, hundreds upon hundreds of individualized Titles. Some are specifically tied to a characters story progression while others get unlocked by filling out the bestiary a certain amount or amassing a lot of money. Since this remaster includes the DLC content, we even get a bunch of costume-specific Titles for each character. Thats in addition to the various costume Titles you unlock through story progression. I wish more RPGs locked cosmetics within a skill or perk tree as this game does with its Titles. It just feels so rewarding and satisfying to unlock, especially because youre earning it through the core gameplay."Titles bestow passive buffs, new skills (or artes as this game terms it), and other unique affects to each character."I mostly prioritized learning new artes when assigning Titles but I quickly learned that levelling up existing arte attacks and stacking up passive buffs to things like critical chance can be much more helpful. This is because combat is legitimately challenging in Tales of Graces. Movement is restricted to a linear axis until you dodge, then you move to the left or right of your target momentarily shifting your position. Perfect Dodges are especially crucial considering how hard bosses can hit. I also like how emphasized weaknesses are in this game. If youre using a strike instead of a slice attack on a slime-like monster, youre going to do peanut damage. Likewise, each particular arte or attack is best suited for a unique enemy type. Battles begin by briefly flashing what monster category each enemy is, and you can assess opponents with the press of a button, making the whole process weakness system frictionless.Party member strategies function the same as other Tales games. When youre not controlling a party member, they behave based on orders you give them in the menus. I find the party AI to be adequate, and fine-tuning them based on a situation typically inches me closer to a victory. That said, the game supports 4-player local co-op for some particularly fun battles. Theres no online co-op support unfortunately, but considering the last couple Tales games didnt have co-op at all, Im not too surprised."Each particular arte or attack is best suited for a unique enemy type."Theres even more gameplay depth and customization with this games version of crafting, which is called Dualizing. Ive heard complaints that Dualizing is unnecessary complex, but its just yet another system to customize your experience with. Like the name suggests, Dualizing allows you to basically combine any consumable, food, equipment, accessory, crafting material, and more with one another. Dualizing equipment with a shard produces levelled-up versions of that equipment. Each shard bestows different buffs, so the combinations can get wild. Likewise, its possible to combine tempered equipments to produce powerful accessories.This doesnt even scratch the surface of Dualizing possibilities within Tales of Graces f. As you can imagine by now, the sheer amount of content stemming from the various systems guarantees hundreds of hours of playtime. I havent even mentioned the wonderful Eleth Mixer and its chance to produces items while on the field. Equipping the Eleth Mixer with Spellbooks confers yet more gameplay affects like converting duplicate enemy drops into Eleth energy.I said Id return to the story, and thats what Ill briefly attempt to do now. The blossoming relationship between Asbel and his friends is believably wholesome, making the twist of the knife hurt that much more when they grow apart and antagonize each other after the games time-skip. The camaraderie between the characters in the intro feels organic, forming a solid springboard for the rest of the story. I admit that I groaned a bit when Sophie was revealed to be an amnesiac who never ages, but the narrative follows through on traditional tropes with solid character development and engaging political machinations despite that. Its the execution that really matters, and in this, Tales of Graces truly excels despite all the well-trod tropes it happily uses."The camaraderie between the characters in the intro feels organic, forming a solid springboard for the rest of the story."Players who dont like the games narrative might still appreciate the understated presentation thereof. Scenes have time to breath, with mood-appropriate silences during key moments and scene music carrying over into urgent battles. Theres a deal of care put into the scene direction, as exemplified by the custom animation rigging for every little character interaction (and theres a lot). The lip synching isnt great and character faces look wooden at times, but considering this game ran off of Wii hardware when it first released, its still impressive. Tales of Graces is surprisingly well-written, capably integrating its characters, themes, and plot together to form a holistic RPG experience that I wont soon forget.The pacing of all the games interlocking systems in tandem with the well-written plot is snappy throughout too. You rarely feel like things are moving at a stagnant pace. You never feel like youre not improving your characters in some way or progressing the plot further. Its a constant stream of improvement and rewards with every step, not dissimilar to the ever-moving gameplay systems in Star Ocean 2: Second Story R from a few years back.Tales of Graces f Remastered is just fantastic all-round. It focuses on a character-rich story and a fun core combat loop without the need to do anything groundbreaking. A story like this has been seen before and the character archetypes are as old as time, but its what the game does with it all that makes it so amazing. Tales of Graces f is a genuine JRPG classic that exudes quality from every corner, and its remastered treatment makes it even easier to love.This game was reviewed on the PlayStation 5.

A SpaceX Falcon 9 rocket lifted off from NASA's Kennedy Space Center on January 15, carrying two lunar landers built independently by private companies Firefly Aerospace and Ispace. NASA / Frank MichauxA SpaceX rocket carrying two commercial lunar landers took off toward space early Wednesday morning.This carpool to the moon is transporting Blue Ghost, a lander built by Texas-based Firefly Aerospace, and Resilience, built by Japanese firm Ispace. The pair launched together on a Falcon 9 rocket from NASAs Kennedy Space Center in Florida at 1:11 a.m. Eastern time. Once the two spacecraft reach the lunar surface, they will conduct independent explorations to collect data and test new technologies.Blue Ghost is carrying ten scientific instruments from NASA in an agreement under the agencys Commercial Lunar Payload Services program. Through this initiative, the agency works with private companies to send lunar devices into space at a lower cost, with the aim to inform NASAs Artemis missions, which plan to return humans to the moon for the first time in more than 50 years.Before we can send our humans back to the moon, we are sending a lot of science and a lot of technology ahead of time to prepare for that, says Nicola Fox, head of NASAs science mission directorate, to CBS News William Harwood.The United States is the only nation that has landed humans on the moon, but five countries have achieved robotic lunar landings: the former Soviet Union, the U.S., China, India and Japan. It wasnt until last year, however, that a private company joined their ranks: Houston-based Intuitive Machines Odysseus spacecraft reached the moons surface in February 2024, though it landed on its side.Now, Blue Ghost and Resilience have the potential to become the second and third private landers to touch down on lunar soil.Blue Ghost is expected to take about 45 days to reach the moon, where it will collect soil samples, test various instruments and monitor X-rays in the Earths magnetic field. Its slated to touch down in Mare Crisium, a volcanic plain. The Blue Ghost lunar lander, as seen at Firefly Aerospace's headquarters in Texas on December 3, 2024 Raquel Natalicchio / Houston Chronicle via Getty ImagesResilience has a longer path ahead. In order to save energy and fuel, it will take four to five months to reach a lunar plain called Mare Frigorisin the moons northern region. Once there, it will collect soil samples, conduct experiments and test technology, such as a water electrolyzer that splits water molecules to form hydrogen and oxygen.This launch marks Ispaces second go at a moon landingits previous landercrashed into the moon in 2023.The newly launched spacecraft will both have a full lunar dayaround two weeks on Earthto conduct their research after landing. Thats because they would not survive the harsh lunar nights, when temperatures can plummet to minus 208 degrees Fahrenheit.Both companies say they have contacted their spacecraft post-launch, and the landers appear to be operating normally, notes CNNs Jackie Wattles.2025 is poised to be a big year for space exploration. Intuitive Machines, in a follow-up to its 2024 success, is set to launch its second lunar mission in February. That lander, called Athena, is intended to take a more direct path to the moon than both Blue Ghost and Resilience, meaning it might beat both of them there, per the New York Times Kenneth Chang.More moon missions are on the schedule for later in the year, and many are meant to help NASA prepare for sending human astronauts to the lunar surface within this decade.This [Blue Ghost] mission embodies the bold spirit of NASAs Artemis campaigna campaign driven by scientific exploration and discovery, says Pam Melroy, NASA deputy administrator, in a statement. Each flight were part of is [a] vital step in the larger blueprint to establish a responsible, sustained human presence at the moon, Mars and beyond.Get the latest stories in your inbox every weekday.Filed Under: Artemis, Innovations, Moon, NASA, Outer Space, Space Travel, SpaceX, Technology

Bonfire Studios, an indie game developer started in 2016, is finally coming out of stealth mode with details on its debut game Arkheron.Read More

Aonic invests $10m in mobile gaming loyalty firm Mega Fortuna"Mega Fortuna is revolutionising how players interact with mobile gaming," says CEO News by Vikki Blake Contributor Published on Jan. 15, 2025 Aonic has made a $10 million minority investment in "mobile gaming loyalty and discovery platform", Mega Fortuna.The investment comes with the option to acquire its full share capital for $70m.Aonic was attracted to Mega Fortuna as its player reward technology could dovetail with other companies in its portfolio."Mega Fortuna is revolutionising how players interact with mobile gaming by seamlessly blending discovery and rewards," said Paul Schempp, Aonic CEO (thanks, VB)."In just two years, they have built an incredible platform that resonates with millions of players worldwide. This investment not only enhances our portfolio but perfectly fits into our Aonic group particularly through significant synergies with Exmox, our leader in performance-based user acquisition."Mega Fortuna CEO, Şeyhmus lker, added: "This milestone is a testament to the exceptional talent, dedication, and innovation of the entire Mega Fortuna team... None of this would have been possible without the incredible teamwork and passion of our people. "Partnering with Aonic and collaborating with Exmox is not just an opportunity to scale our vision but also a recognition of the outstanding effort that brought us here. I couldnt be prouder of what weve achieved together."

Justin Carter, Contributing EditorJanuary 15, 20252 Min ReadImage via Sony.At a GlanceSony has filed multiple new patents that could shape the future of its controllers.Several patents filed by Sony (spotted by outlets like GameSpot and 80LV), and offer some potential insight into PlayStation's technological future.One application filed in June 2024 concerns a machine learning-based model that appears to address "input lag," the time passed between a pressed controller and the action occuring onscreen. The model is "configured to release user commands to a computer game at a time when the command correctly aligns with what is going on in[it,]" according to the patent.The patent's summary notes input lag is a problem for both system engineers and players, and believes machine learning could "control the computer game according to the user command in advance of the user completing the user command based on the identification."To achieve this, Sony is considering a sensor with a camera that captures the player and their controller, which is then used to train the machine learning program. The technology seems to primarily concern online multiplayer games, where input lag has a dramatic effect.Similarly, one application from June 2023 focuses on machine learning for predicting user commands "before the input action is completed, or might be delayed after completion and processing until a later time when the command is most appropriate for release."Another notable patent centers on a physical attachment that would let players hold the DualSense controller like a firearm. The attachment's apparent aim is to be "more intuitive" for players, as the description notes some players "may lack the dexterity/motor skills" to conventionally hold a controller, and hopes the attachment "alleviates or mitigates the issue."Sony previously experimented with gun-themed control schemes with two shooting attachments for the PlayStation Move, which were officially supported by a number of PlayStation 3 titles.Filing a patent doesn't mean Sony will bring any of its proposed technology to market. In the past, Sony filed patents for technology such as NFTs and local multiplayer for live, in-person events. To date, nothing has come of either, and it quickly shot down speculation that its PlayStation Stars loyalty program was an NFT platform in disguise.You can see Sony's list of published patents here. Among other things, the more recent ones concern inputs, image upscaling, and using machine learning for online multiplayer.Read more about:[Company] PlayStationAbout the AuthorJustin CarterContributing Editor, GameDeveloper.comA Kansas City, MO native, Justin Carter has written for numerous sites including IGN, Polygon, and SyFy Wire. In addition to Game Developer, his writing can be found at io9 over on Gizmodo. Don't ask him about how much gum he's had, because the answer will be more than he's willing to admit.See more from Justin CarterDaily news, dev blogs, and stories from Game Developer straight to your inboxStay UpdatedYou May Also Like

Following a scrubbed launch attempt and weather-related delays, Blue Origin will once again try to send its New Glenn rocket into space for the first time. During the attempt, the Jeff Bezos-owned space company aims to reach orbit, helping to further its goals of shuttling Project Kuiper satellites, equipment, and eventually humans into space.Heres what you need to know about when and how to watch New Glenns long-awaited inaugural launch.What is New Glenn?New Glenn is the 320-foot-tall rocket that Blue Origin initially announced in 2016. Though Blue Origin planned to launch New Glenn by 2020, the project was beset by delays due to issues with engine development and other technical setbacks.The rocket has a reusable first stage powered by the companys BE-4 engines, which run on liquified natural gas and liquid oxygen. Shortly after launch, the first stage is supposed to detach and autonomously land upright on a sea-based platform, where Blue Origin can then retrieve it and reuse it for future missions.Following separation, New Glenns upper stage should fire up its BE-3U engines a less powerful engine that uses liquid oxygen and liquid hydrogen as it attempts to propel itself into space with its payload. The upper stage is capable of delivering 45 metric tons of cargo into low Earth orbit.For this uncrewed launch, New Glenn will house the Blue Ring Pathfinder, a payload consisting of a communications array, a power system, and a flight computer. This will let Blue Origin test its Blue Ring spacecraft, which will eventually support missions with refueling, hosting, data relay, and cloud computing capabilities.New Glenns first stage uses BE-4 engines, while its upper stage has less powerful BE-3U engines. Image: Blue OriginEarlier this week, Blue Origin scrubbed New Glenns launch due to a vehicle subsystem issue. A successful first launch could make Blue Origin a serious rival to Elon Musks SpaceX. Along with competing for government contracts, both commercial space companies are also working to build out internet satellite constellations, with SpaceX regularly sending Starlink satellites into space and Blue Origin on tap to support Amazons Project Kuiper initiative.When will Blue Origin launch New Glenn?Blue Origins next launch attempt will take place at Launch Complex 36 in Cape Canaveral, Florida, on Thursday, January 16th. The three-hour launch window opens at 1AM ET (10PM PT). How to watch New Glenns launchYou can watch a livestream of New Glenns launch from Blue Origins website, its X account, and its YouTube channel. Well embed a livestream here when it becomes available.

Large Language Models (LLMs) and Vision-Language Models (VLMs) transform natural language understanding, multimodal integration, and complex reasoning tasks. Yet, one critical limitation remains: current models cannot efficiently handle extremely large contexts. This challenge has prompted researchers to explore new methods and architectures to improve these models scalability, efficiency, and performance.Existing models typically support token context lengths between 32,000 and 256,000, which limits their ability to handle scenarios requiring larger context windows, such as extended programming instructions or multi-step reasoning tasks. Increasing context sizes is computationally expensive due to the quadratic complexity of traditional softmax attention mechanisms. Researchers have explored alternative attention methods, such as sparse attention, linear attention, and state-space models, to address these challenges, but large-scale implementation remains limited.Sparse attention focuses on relevant inputs to reduce computational overhead, while linear attention simplifies the attention matrix for scalability. However, adoption has been slow due to compatibility issues with existing architectures and suboptimal real-world performance. For example, state-space models effectively process long sequences but often lack the robustness and accuracy of transformer-based systems in complex tasks.Researchers from MiniMax have introduced the MiniMax-01 series, including two variants to address these limitations:MiniMax-Text-01: MiniMax-Text-01 comprises 456 billion total parameters, with 45.9 billion activated per token. It leverages a hybrid attention mechanism for efficient long-context processing. Its context window extends to 1 million tokens during training and 4 million tokens during inference.MiniMax-VL-01:The models employ a novel lightning attention mechanism, reducing the computational complexity of processing long sequences. Also, integrating a Mixture of Experts (MoE) architecture enhances scalability and efficiency. The MiniMax models feature 456 billion parameters, of which 45.9 billion are activated for each token. This combination allows the models to process context windows of up to 1 million tokens during training and extrapolate to 4 million tokens during inference. By leveraging advanced computational strategies, the MiniMax-01 series offers unprecedented capabilities in long-context processing while maintaining performance on par with state-of-the-art models such as GPT-4 and Claude-3.5.The lightning attention mechanism achieves linear computational complexity, enabling the model to scale effectively. The hybrid attention architecture alternates between lightning and softmax attention layers, ensuring a balance between computational efficiency and retrieval capabilities. The models also incorporate an enhanced Linear Attention Sequence Parallelism (LASP+) algorithm, efficiently handling extensive sequences. Also, the vision-language model MiniMax-VL-01 integrates a lightweight vision transformer module, enabling it to process 512 billion vision-language tokens through a four-stage training process. These innovations are complemented by optimized CUDA kernels and parallelization strategies, achieving over 75% Model Flops Utilization on Nvidia H20 GPUs.Performance evaluations reveal that the MiniMax models achieve groundbreaking results across various benchmarks:For instance, MiniMax-Text-01 is 88.5% accurate on MMLU and performs competitively against models like GPT-4.The vision-language model MiniMax-VL-01 surpasses many peers, with a 96.4% accuracy rate on DocVQA and 91.7% on AI2D benchmarks.These models also offer a 2032 times longer context window than traditional counterparts, significantly enhancing their utility for long-context applications.In conclusion, the MiniMax-01 series, comprising MiniMax-Text-01 and MiniMax-VL-01, represents a breakthrough in addressing scalability and long-context challenges. It combines innovative techniques like lightning attention with a hybrid architecture. By leveraging advanced computational frameworks and optimization strategies, researchers have introduced a solution that extends context capabilities to an unprecedented 4 million tokens and matches or surpasses the performance of leading models like GPT-4.Check out the Paper and Models on Hugging Face. All credit for this research goes to the researchers of this project. Also,dont forget to follow us onTwitter and join ourTelegram Channel andLinkedIn Group. Dont Forget to join our65k+ ML SubReddit.(Promoted) Asif RazzaqAsif Razzaq is the CEO of Marktechpost Media Inc.. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences. Meet 'Height':The only autonomous project management tool (Sponsored)

LatestMachine LearningRevolutionizing AI with Jamba: The Cost-Effective Game-Changer for Long Contexts 0 like January 15, 2025Share this postAuthor(s): Rohit Sharma Originally published on Towards AI. If you think all LLMs are the same think again. Every time I find something new when I deep dive into a new framework!Ive been of late experimenting with Jamba and as a GenAI architect whos tested it extensively Ive been blown away by what it can achieve and would make us re-think our solutions going forward.All This while simplifying workflows and slashing the costs!Lets dive into why this model is making waves.Jamba isnt yet another name in the crowded AI landscape its a breakthrough model thats redefining the science of how we approach the long-context tasks, cost-efficiency and GenAI architectures. From ingesting entire annual reports in a single shot to natively supporting tool-calling for agentic apps.Core Abilities1. Real Long Context Length: Beyond RAG Without a Vector DBWhat it does: Jamba eliminates the need for a vector DB in many cases because of its ability to handle massive docs directly in its 256K context window. This removes the need for chunking, embedding and retrieval pipelines.Why it matters: Unlike many models Jambas claimed/promised context length aligns with its actual performance. During testing, I loaded an entire annual report into the context, and Jamba processed it with 85% accuracy on insight extraction tasks. Run-time inclusion of documents in RAG workflows is going to be the biggest use case here. Long document summarization and insight extraction. Analyzing call transcripts or long chat histories. Multi-hop reasoning in agentic systems.2. Out-of-the-Box Conversational RAGWhat it does: Jamba has native support for RAG that takes care of chat history, chunking, indexing, and retrieval strategies, making it ideal for conversational AI applications.Why it matters: GenAI architects can leverage these capabilities without building custom RAG pipelines unless the use cases or the documents complexity demands it. This would accelerate deployment. I see this as a huge help in Building intelligent customer support bots that have a dynamic ever changing document knowledge base. Context-aware multi-turn conversations in enterprise chat tools. All of this was possible anyway but the velocity of the solution development is going to be 10xed (for certain use-cases as I said).3. Enhanced RAG PipelinesWhat it does: Even in traditional RAG workflows/pipelines involving Vector DBs Jambas ability to of handling massive context lengths would improve the final synthesis due to inclusion of complete context. This would be particularly useful for solutions where the context length of the retrieved documents used to be limited by the LLMs promised context length. And lets face it most of the times the actual-context-length never matches the promised-context-length when one starts comparing the synthesis quality of the final response.Why it matters: Longer context capabilities enable handling larger document batches and multi-turn chat histories enhancing quality. Legal/medical/compliance workflows with large knowledge management systems requiring high recall rate are going to benefit from this a lot.4. Agentic App ReadinessWhat it does: Jamba supports native tool-calling alongside its long-context abilities which makes it an ideal model for agentic applications and complex reasoning tasks (at lower cost and lightweight architecture).Why it matters: The ability to natively invoke external keeps the doors open for dynamic and interactive agentic workflows. I see a huge value of this in advanced reasoning agents in operational workflows and financial analysis that require real-time API integration.5. Output FormattingWhat it does: Jamba supports native JSON output formatting, streamlining integration with downstream systems.Why it matters: Structured outputs reduce parsing errors and improve automation.Cost and Efficiency1. Efficiency GainsJamba delivers 3x throughput on long contexts compared to similar models, like Mixtral while maintaining accuracy.Its hybrid architecture combining Mamba (SSM) and Transformer layers optimizes compute usage for high performance.2. Lower CostsEliminates the need for VDBs in static workflows, reducing infrastructure costs.Fits 140K tokens on a single GPU, minimizing hardware requirements.3. Optimized Latency and ThroughputAchieves faster response times, even with large input contexts, enabling real-time use cases.Simplifying ArchitecturesJambas unique long-context handling enables simpler, more streamlined architectures:Without Vector Databases: Ingest documents directly into the prompt for static use cases like annual reports or legal contracts. Reduce the architectural overhead of embedding, chunking, and retrieval pipelines.Streamlined RAG Pipelines: Handle larger, more relevant document batches with fewer retrieval operations.Examples:Legal Analysis: Process contracts without retrieval systems, answering queries directly from the document.Customer Support: Load product manuals or FAQs directly into context for instant, context-aware responses.Compliance Audits: Analyze policy documents or regulations in a single pass, reducing pre-processing overhead.Comparing Jamba with Other ModelsHere is a quick comparison of Jamba with popular LLMs in market (Source)Final TakeawaysJamba has the potential of redefining some GenAI specific workflows by enabling real long-context handling, involving lightweight architectures and reducing costs.Its unique combination of long context lengths, native tool-calling, and efficient compute usage makes it an excellent option for GenAI architects.Whether youre analyzing massive documents, running agentic systems, or building cost-sensitive AI solutions Jamba is worth exploring.Ready to dive in? Jamba is live on Hugging Face (Links below)Key Links:Jamba: https://www.ai21.com/jambaModel Cards:ai21labs/AI21-Jamba-1.5-Large: https://huggingface.co/ai21labs/AI21-Jamba-1.5-Largeai21labs/AI21-Jamba-1.5-Mini: https://huggingface.co/ai21labs/AI21-Jamba-1.5-MiniJoin thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming asponsor. Published via Towards AITowards AI - Medium Share this post

Marvel Rivals players are being treated to an unlikely crossover that brings the Advanced Suit 2.0 from Marvels Spider-Man 2 into the game as a new skin.PlayStation announced the news itself in a post on X/Twitter, revealing a first look at how NetEase Games is reimagining the classic video game outfit for its new hero shooter.This sleek Spider-Man suit made its debut as an original design for Insomniac Games Marvels Spider-Man, going on to make appearances in all three entries in that series. The main element setting this suit apart from the rest is its now-iconic white spider symbol, with its appearance in Marvel Rivals cementing it as a surprise collaboration with Sony and its console-exclusive superhero game. Expect Spider-Mans Advanced Suit 2.0 to swing into the in-game store on January 30, the same day Marvels Spider-Man 2 comes to PC.Marvel Rivals has a surprise collaboration with Sony's Spider-Man game.Marvel Rivals launched in December and already offers a smorgasbord of unlockable suits, but this new Spidey suit arrives as an especially unique addition. Along with its ties to the massively popular PlayStation series, the Advanced Suit 2.0 will once again be worn by a Spider-Man who is voiced by Yuri Lowenthal. The actor has played Peter Parker in all three of Insomniacs Spider-Man games and also happens to have provided his talents for Marvel Rivals take on the character. In other words, those who pick up the costume come January 30 should feel right at home as they thwip their way to victory.NetEase's decision to bring Spider-Mans Advanced Suit 2.0 into the fold gives players another piece of content to look forward to among everything that dropped with Season 1: Eternal Night Falls last week. This first season of content added Fantastic Four representatives Mister Fantastic and the Invisible Woman as new playable characters, with The Thing and Human Torch set to join in just a few weeks. Much more is on the way, too, as creative director Guangyun Chen recently promised to launch at least one new hero every month and a half.While we wait to try out that new Spider-Man suit for ourselves, be sure to read up on some of the custom skins players have modded into the game. You can also learn about all of the balance changes introduced with Season 1 and how some players are using an Invisible Woman ability to detect what they believe are bot players.Michael Cripe is a freelance contributor with IGN. He started writing in the industry in 2017 and is best known for his work at outlets such as The Pitch, The Escapist, OnlySP, and Gameranx.Be sure to give him a follow on Twitter @MikeCripe.